Abstract
We have developed a false face reduction algorithm dedicated for face detection in an unconstrained environment, based on the discriminative facial T-shape region constructed by human eyes, nose, and mouth. The algorithm started with a detected face, through lighting compensation, normalization, facial T-shape region extraction, and eigenface selection using genetic algorithms to classify face/false face robustly. Results of face detection with and without the proposed false face reducer show substantial improvements on precision rate at a little loss of recall rate.
